This summer, Thomson Reuters introduced a generative AI reinvigoration of the platform, adding AI-powered functions to Westlaw’s legal software with CoCounsel Legal. The generative AI additions, which are slowly rolling out to all users, essentially act as an AI legal assistant capable of research, drafting, document analysis, documentation and search. Chief Product Officer David Wong said it has been trained to conduct research much like a lawyer or professional researcher, and has agentic capabilities to perform complex multi-step tasks, such as analyzing contracts or preparing for a deposition. The AI upgrades can save an attorney six to 10 hours of work, Wong estimated.

This update has been long in the making. Through tech upgrades, acquisitions, user feedback and careful attention to user needs and behavior, Thomson Reuters has been developing these generative-AI-powered features for years. I spoke with Wong, CEO Steve Hasker and Chief Technology Officer Joel Hron about their process of bringing the power of generative AI to Westlaw. Today’s Forbes CEO newsletter focuses on our discussion about how a company can reinvent and improve what it already has on the market with AI. 

If you like what you read here, you can easily share it online and on your social media pages. Today’s newsletter, and many other recent editions of Forbes CEO, can be found on our website here.

Hasker said that Thomson Reuters has always paid attention both to technology and Westlaw customer needs, but about five years ago they started what they called the Change Program, which aimed to re-engineer customer experience and revamp Westlaw’s entire tech stack, moving things to the cloud. The timing of this technology upgrade made Westlaw’s system and data ready to work with generative AI early on.

“Sometimes, it’s better to be lucky than good, insofar as we got through that work just in time for this gen AI-driven revolution,” Hasker said.

But, Chief Technology Officer Joel Hron said, the very nature of Westlaw’s setup also made it ripe to add generative AI. Customers already used the service for legal research and to perform sophisticated tasks. The data about case law, statutes, torts and legal writing was already in the system.

“It’s not like we had to go rebuild all of Westlaw,” Hron said. “We were able to leverage the same search algorithms and the same sort of capabilities that underpin Westlaw. It just synthesized that information in a different way for users with generative AI.”

When ChatGPT passed the bar exam, which OpenAI reported in 2023, Hasker said it became clear that generative AI was an important tool for lawyers—and something Thomson Reuters needed to bring to Westlaw.

“Sometimes the constraint is that even though you might have the strategy or the will or the desire to build something, you might not have the talent yet,” Wong said. “And as we all know, hiring experienced AI professionals, it’s just hard. It takes time.” 

The lack of bandwidth and talent led Thomson Reuters to make acquisitions in the nascent space of AI for legal research work. Wong said his team began researching who was doing the types of things Thomson Reuters wanted to do with Westlaw, as well as who had access to the leading technology. 

Hasker said the company has spent about $2.5 billion on AI-related acquisitions to date, and has another $10 billion available.

In 2022, they acquired ThoughtTrace—which led to Hron joining the company—bringing natural language processing to contracts. In 2023, they bought Casetext—which had been one of the first companies to work with OpenAI and was the initial creator of CoCounsel—for $650 million. Last year, they bought legal LLM company Safe Sign Technologies, as well as Materia, a company that specializes in agentic AI for applications in accounting, taxes and auditing.

Wong and Hasker said that when making this kind of acquisition, they look for companies that will enhance Thomson Reuters’ culture by bringing new questions and perspectives to the table to improve products. They also have looked for companies that will quickly be a boon to Thomson Reuters as a whole, with no significant tech debt and offerings that can easily add value to existing products—and shareholders.

With the speedy evolution of AI technology, Hron said that durability is also important. What is built today needs to still be workable next week, next month and next year. General tech expertise on the bleeding edge of what’s new helps here, but so does domain expertise: in this case, legal knowledge.

“The models really don’t function well without the context of the business,” he said. “I think businesses and companies have that as well are a really strong way to build durability.”

For all employees, Hasker said, Thomson Reuters has added some base-level AI training and knowhow. 

“There’s no way to fudge that,” Hasker said. “You can’t sort of delegate that to other people and you can’t delegate that to technology. You actually have to personally invest in it, and then follow through.”

The employees working with AI, however, need to be proficient in it at a much higher level, which Wong said is a differentiator for Thomson Reuters’ AI projects. There are a couple of projects the company has ongoing to get more employees into that caliber of AI use. Thomson Reuters has long had a division called Thomson Reuters Labs, which focuses on applied research in cutting-edge technology. In the last couple of years, that team has more than doubled in size, and most of the work has pivoted to improve Thomson Reuters’ own products.

The company was also quick to establish its own generative AI “playground.” Called Open Arena, Hron said it was first created in the months after OpenAI’s ChatGPT launched. It’s a protected place to learn and experiment with AI, and Hron noted that about 17,000 of the company’s employees are currently using it.

Thomson Reuters executives recognize that this is a real problem for any AI legal work.

“There is no such thing as hallucination-free,” Hron said. “It’s all about how you spin that fly wheel of verification.”

Hron noted that the AI features on CoCounsel are built with extra rigor and testing to minimize hallucination, incorporating additional teams and processes hidden from the user to ensure that almost everything appearing on the platform is truthful. But they’ve also been designing a user experience for Deep Research that provides easy verification. Case citations appear highlighted in blue if they actually go somewhere, and the user can easily cross-reference that case. If AI has invented a case, there is no blue highlight—and clicking on it won’t take the user anywhere. Other visual cues, Hron said, should make verification apparent.

“What you’ll see is as we do more of this throughout CoCounsel, UX will become more about giving cues to humans to do verification versus putting buttons in places for humans to take action,” Hron said.

Hron said they’re also building in a tool to let users decide how much leeway the AI can have for different projects. It can have more freedom for more creative applications, but for tasks that require precision, AI’s research influence can be minimized.

“Our aspiration is to play a much broader and more important role in the success of our customers,” Hasker said.

As more AI-powered functions for CoCounsel Legal roll out now, Deep Research in Practical Law—a new guidance module that helps plan research and lead to a deep and nuanced analysis—is in beta right now. The company plans to fully roll it out in 2026.

Thomson Reuters is also planning to bring AI to its tax, accounting and audit platforms, Wong said. These capabilities will integrate with CoCounsel for financial and tax law, but they will also be able to prepare complex tax filing documents. Wong said that this could help an accountant work with different customers by using their financial documents and AI reasoning to figure out the impact of different financial choices—like with retirement accounts, savings or investment properties. It could also provide a significant amount of automation in the CFO’s office, freeing up more time for the finance team to work on strategy.

Wong said that the new focus on AI is helping drive a shift that’s already been in place at Thomson Reuters, which was once chiefly known for newspaper publishing and its news wire service.

“Internally, I think we’ll become much more of a technology company,” Wong said. “Steve had said that we’ve been on a journey from a content and media company into a content-driven technology company. I think we’re on that path. I’d say maybe we’re halfway there, but I think that in another five years, we will be indistinguishable from the leading software companies in the world.”

“The No. 1 skill we look for, regardless of tech or anything, is adaptability,” Hron said. “I think the ability for people to be curious and learn new things and fail quickly is the most critical thing for any company right now.”

Wong added that companies must know what problem AI will solve for them. If AI is implemented just for the sake of AI—like creating a chatbot without stating its purpose—it won’t go far. Even more mundane aims—like using AI to improve a customer service call center system—could be a big deal. After all, companies in competitive consumer marketplaces can use their customer service to differentiate.

Hron said that AI also lets companies try to solve big problems first. Instead of starting small and working their way up to bigger problems, AI makes it easy to pilot solutions to the biggest problems. Once a big solution is found, Hron said he then starts focusing more on the smaller details.

“I think starting big and then focusing in lead you down a better path,” he said.”

Everyday employees who go above and beyond with their contributions to the company deserve recognition. But it’s important to ensure that employees see the recognition as positive. All kinds of praise don’t work for all employees, and what you say and how you say it speaks volumes.